xgbe: use standard rounddown_pow_of_two

[dougm]

Summary

This file has an implementation of __rounddown_pow_of_two which is essentially the same as rounddown_pow_of_two in libkern, except written in a way that triggers coverity warnings:

CID 114616:  Integer handling issues  (INTEGER_OVERFLOW)
Expression "fls_long(n) - 1U", which is equal to 4294967295, where "fls_long(n)" is known to be equal to 0, underflows the type that receives it, an unsigned integer 32 bits wide.

Use the standard implementation.

[markj]

Assuming it came directly from the commit log message, the title should start with "axgbe:".